[FBumann]

Prompt: 2. Calls: subject[property] in sum, at, within= and conditions, the presence condition, and the inner join for several properties.

Note

The following content was generated by AI.

What this changes

sum, at, within= and conditions now read subject[property, …], by the join and group rules of a relation column. One property reads its own table, subject.property. Several properties read the inner join of their tables. subject[property] alone tests that a row has the property. Stacked on #1.

The PR also fixes one bug. position(..., within=...) in a condition named no table in names_read, for relations too, so a consumer that attaches what a mask reads left the groups out.

Why

This is PR 2 of the stack in which sets and properties replace relations.

How a read lowers

Namespace.table(subject, names) decides which table a call reads:

written	program node's name	tables the data supplies
sum(x, over=generator, by=generator[bus])	generator.bus	generator.bus
sum(x, over=generator, by=generator[bus, carrier])	generator (the inner join)	generator.bus, generator.carrier
sum(x, over=start, by=cover[covered])	cover	cover
within=snapshot[period]	snapshot.period	snapshot.period
generator[bus] == 'b1'	RelationComparison('generator.bus', 'bus', …)	generator.bus
line[bus0] != line[bus1]	RelationPairComparison('line.bus0', …, 'line.bus1', …)	both
generator[bus, carrier] alone	RelationDefined on each table, joined by And	both

• The inner join is no table of the program. For several properties, the node carries a Relation with subject set, keyed by the subject's members, with one value column per property. Its missing is absent where any of the properties is.
• JoinColumns.tables and Partition.tables name the tables the data supplies. Mask.names_read reads them, for a join, a partition and a position counted within groups, so a mask never names the subject as if it were data.
• JoinColumns.written and Partition.written give the subject as the file wrote it. The dimension-rule messages use it.
• The typesetter takes each shape from the node, so it never looks up a view in program.relations.
• Conditions: the grammar parses name[a, …] alone as a presence test. subject[a, b] == x keeps the rewrite "a comparison reads one column", which now fires at the comparator.
• Messages name the subject as written. For example: sum(by=generator[zone]) names ['zone'], which is no property of 'generator'. Name one of its properties, ['bus', 'carrier'].

Departures, and what this PR does not do

• The position(within=) fix rides this PR rather than a PR of its own. It also changes behaviour for relations. I first reproduced it as a strict xfail in three cases (a relation, one property, two properties) and watched all three fail. The marker came off with the fix.
• A test changed its expected wording. The test that asserts Only a relation has columns (tests/test_validation.py) now expects Only a relation, a set or a dimension has columns, because a dimension and a set now take brackets.
• No typesetting work. The math prints the table names as they are, such as generator.bus(g), and the reference page shows that output. PR 4 changes the legend and regenerates both.
• No data contract. docs/reference/reading.md is PR 4.

Gates, and the mutation table

• Passed: pixi run lint, pixi run test (3202 passed, 1 skipped) and pixi run docs-build (strict). Both YAML examples on the page load, and the math and the quoted refusal were taken from the code's output.
• Not run: compile-tex. Tectonic cannot download its bundle in this container.
• No CI run. GitHub Actions has never run in this fork, and dispatching ci.yml returns 404. Actions must be turned on in the fork's settings.
• Mutation table: the same script as feat(language): a file may declare sets of labels and the properties of a dimension or a set #1 took it. It refuses a dirty tree, restores each file with git checkout --, and drops __pycache__ around each mutation. The tree came back clean.
  • In the first run, one deletion stayed green: the guard on presence for a name that is not a subject. Without it, the file was still refused, but with the wrong message. The guard also called a declared parameter "not found", which is now fixed.
  • Two probes now cover it. The second row in the table is the guard deleted again by hand after the fix: both probes fail.

guard deleted	suite
a dimension read by no property	caught
a column the subject lacks	caught (5)
a dimension with properties takes brackets	caught (15)
one property reads its own table	caught (4)
the inner join is absent where a property is	caught
presence of a relation is refused	caught
presence of a name that is not a subject	green in the first run → 2 probes added, caught (2)
presence of a property the subject lacks	caught
a comparison reads the property table	caught (3)
a comparison of a column the subject lacks	caught (2)
a subject with a column is read as a table	caught (4)
a column of a non-subject is refused	caught (2)
presence parses	caught (8)
the one-column rewrite after a listed read	caught (2)
a view reads the tables of its properties	caught (4)
a joined condition names the tables it reads	caught
a dim refusal names the subject as written	caught
the typesetter reads the shape off the join	caught (3)
a position names the tables of its groups	xfail reproduced first, 3 cases

🤖 Generated with Claude Code

https://claude.ai/code/session_01W6yvGzZN2iR4WyA9RumvvJ

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}